Re: [PATCH 2.6.13.1] Patch for invisible threads

[Al Viro <viro@ZenIV.linux.org.uk>]

On Tue, Sep 13, 2005 at 07:53:40AM -0700, Linus Torvalds wrote:

> So this patch is _wrong_.

Definitely.
 
> I think the problem is "proc_check_root()", which just refuses to do a lot 
> of things without a fs. Many of those things are unnecessary, afaik - we 
> should allow it. But allowing it means that some other paths may need more 
> checking..
>
> So you can _try_ to just make proc_check_root() return 0 when 
> proc_root_link() returns an error...

I very much doubt the correctness of that.

The real problem here is obvious: it's about permissions on /proc/<pid>/task.
That's where the things go wrong - we use proc_permission() for it and we
have group leader as associated task.

Note that stuff _in_ proc/<pid>/task will keep working just fine, if we
manage to get to it - there we have other threads as associated tasks,
so everything works as it should.

What we need is to decide what kind of access control do we really want on
/proc/<pid>/task.  That's it.